diff --git a/.changeset/demote-appium-uiautomator2-driver.md b/.changeset/demote-appium-uiautomator2-driver.md
deleted file mode 100644
index eb04073a9..000000000
--- a/.changeset/demote-appium-uiautomator2-driver.md
+++ /dev/null
@@ -1,11 +0,0 @@
----
-"@qawolf/cli": patch
----
-
-move appium-uiautomator2-driver to devDependencies and remove it from the managed runtime
-
-The runtime does not load this driver from node_modules. Appium loads it from APPIUM_HOME, where `qawolf install android` installs it.
-
-The devDependency stays so the build can read the pinned version for the APPIUM_HOME install.
-
-This prevents lockfile churn in consumer repositories when a future driver version ships a problematic npm-shrinkwrap.json.
diff --git a/.changeset/tidy-skills-explain.md b/.changeset/tidy-skills-explain.md
deleted file mode 100644
index fa84e1948..000000000
--- a/.changeset/tidy-skills-explain.md
+++ /dev/null
@@ -1,7 +0,0 @@
----
-"@qawolf/cli": patch
----
-
-Teach the shipped `qawolf-cli` skill the runner group, written for a harness that has a shell, an API key and its own vision model and none of QA Wolf's tooling.
-
-The guide is a skill resource rather than part of the skill body, so it costs nothing until a runner is what the task needs. It leads with what is expensive to discover: relaunching an id attaches rather than billing a second pod; a runner launched for you has a fresh browser with nothing signed in and no page open; only a run starts the screen, so a screenshot before one fails however long you wait; the recorder carries the `locator` and `alternates` a screenshot cannot give you; `exec` returns whether a snippet ran and not its value, so results come back through the `console` stream; and an unreachable answer from `qawolf runner run` does not mean the run did not start, so resubmitting bills a second one. Auth and the reads-versus-writes framing account for the group as well.
diff --git a/CHANGELOG.md b/CHANGELOG.md
index fe1524e82..5b4d60943 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,21 @@
 # @qawolf/cli
 
+## 1.9.2
+
+### Patch Changes
+
+- 62d07c4: move appium-uiautomator2-driver to devDependencies and remove it from the managed runtime
+
+  The runtime does not load this driver from node_modules. Appium loads it from APPIUM_HOME, where `qawolf install android` installs it.
+
+  The devDependency stays so the build can read the pinned version for the APPIUM_HOME install.
+
+  This prevents lockfile churn in consumer repositories when a future driver version ships a problematic npm-shrinkwrap.json.
+
+- a442520: Teach the shipped `qawolf-cli` skill the runner group, written for a harness that has a shell, an API key and its own vision model and none of QA Wolf's tooling.
+
+  The guide is a skill resource rather than part of the skill body, so it costs nothing until a runner is what the task needs. It leads with what is expensive to discover: relaunching an id attaches rather than billing a second pod; a runner launched for you has a fresh browser with nothing signed in and no page open; only a run starts the screen, so a screenshot before one fails however long you wait; the recorder carries the `locator` and `alternates` a screenshot cannot give you; `exec` returns whether a snippet ran and not its value, so results come back through the `console` stream; and an unreachable answer from `qawolf runner run` does not mean the run did not start, so resubmitting bills a second one. Auth and the reads-versus-writes framing account for the group as well.
+
 ## 1.9.1
 
 ### Patch Changes
diff --git a/package.json b/package.json
index 9f8a4451e..d2057f7b8 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@qawolf/cli",
-  "version": "1.9.1",
+  "version": "1.9.2",
   "description": "Run and manage QA Wolf flows from the terminal, CI, or an AI agent",
   "keywords": [
     "automation",